src/gromacs/gmxlib/nonbonded/nb_kernel_avx_256_single/nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_avx_256_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013,2014, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS avx_256_single kernel generator.
  37  */
  38 #include "config.h"
  39
  40 #include <math.h>
  41
  42 #include "../nb_kernel.h"
  43 #include "gromacs/legacyheaders/types/simple.h"
  44 #include "gromacs/math/vec.h"
  45 #include "gromacs/legacyheaders/nrnb.h"
  46
  47 #include "gromacs/simd/math_x86_avx_256_single.h"
  48 #include "kernelutil_x86_avx_256_single.h"
  49
  50 /*
  51  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_avx_256_single
  52  * Electrostatics interaction: Ewald
  53  * VdW interaction:            LennardJones
  54  * Geometry:                   Particle-Particle
  55  * Calculate force/pot:        PotentialAndForce
  56  */
  57 void
  58 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_VF_avx_256_single
  59                     (t_nblist                    * gmx_restrict       nlist,
  60                      rvec                        * gmx_restrict          xx,
  61                      rvec                        * gmx_restrict          ff,
  62                      t_forcerec                  * gmx_restrict          fr,
  63                      t_mdatoms                   * gmx_restrict     mdatoms,
  64                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  65                      t_nrnb                      * gmx_restrict        nrnb)
  66 {
  67     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  68      * just 0 for non-waters.
  69      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
  70      * jnr indices corresponding to data put in the four positions in the SIMD register.
  71      */
  72     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  73     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  74     int              jnrA,jnrB,jnrC,jnrD;
  75     int              jnrE,jnrF,jnrG,jnrH;
  76     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  77     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
  80     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  81     real             rcutoff_scalar;
  82     real             *shiftvec,*fshift,*x,*f;
  83     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
  84     real             scratch[4*DIM];
  85     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  86     real *           vdwioffsetptr0;
  87     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  88     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
  89     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  90     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  91     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
  92     real             *charge;
  93     int              nvdwtype;
  94     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
  95     int              *vdwtype;
  96     real             *vdwparam;
  97     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
  98     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
  99     __m256i          ewitab;
 100     __m128i          ewitab_lo,ewitab_hi;
 101     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 102     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 103     real             *ewtab;
 104     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 105     real             rswitch_scalar,d_scalar;
 106     __m256           dummy_mask,cutoff_mask;
 107     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 108     __m256           one     = _mm256_set1_ps(1.0);
 109     __m256           two     = _mm256_set1_ps(2.0);
 110     x                = xx[0];
 111     f                = ff[0];
 112
 113     nri              = nlist->nri;
 114     iinr             = nlist->iinr;
 115     jindex           = nlist->jindex;
 116     jjnr             = nlist->jjnr;
 117     shiftidx         = nlist->shift;
 118     gid              = nlist->gid;
 119     shiftvec         = fr->shift_vec[0];
 120     fshift           = fr->fshift[0];
 121     facel            = _mm256_set1_ps(fr->epsfac);
 122     charge           = mdatoms->chargeA;
 123     nvdwtype         = fr->ntype;
 124     vdwparam         = fr->nbfp;
 125     vdwtype          = mdatoms->typeA;
 126
 127     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 128     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 129     beta2            = _mm256_mul_ps(beta,beta);
 130     beta3            = _mm256_mul_ps(beta,beta2);
 131
 132     ewtab            = fr->ic->tabq_coul_FDV0;
 133     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 134     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 135
 136     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 137     rcutoff_scalar   = fr->rcoulomb;
 138     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 139     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 140
 141     rswitch_scalar   = fr->rcoulomb_switch;
 142     rswitch          = _mm256_set1_ps(rswitch_scalar);
 143     /* Setup switch parameters */
 144     d_scalar         = rcutoff_scalar-rswitch_scalar;
 145     d                = _mm256_set1_ps(d_scalar);
 146     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 147     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 148     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 149     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 150     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 151     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 152
 153     /* Avoid stupid compiler warnings */
 154     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 155     j_coord_offsetA = 0;
 156     j_coord_offsetB = 0;
 157     j_coord_offsetC = 0;
 158     j_coord_offsetD = 0;
 159     j_coord_offsetE = 0;
 160     j_coord_offsetF = 0;
 161     j_coord_offsetG = 0;
 162     j_coord_offsetH = 0;
 163
 164     outeriter        = 0;
 165     inneriter        = 0;
 166
 167     for(iidx=0;iidx<4*DIM;iidx++)
 168     {
 169         scratch[iidx] = 0.0;
 170     }
 171
 172     /* Start outer loop over neighborlists */
 173     for(iidx=0; iidx<nri; iidx++)
 174     {
 175         /* Load shift vector for this list */
 176         i_shift_offset   = DIM*shiftidx[iidx];
 177
 178         /* Load limits for loop over neighbors */
 179         j_index_start    = jindex[iidx];
 180         j_index_end      = jindex[iidx+1];
 181
 182         /* Get outer coordinate index */
 183         inr              = iinr[iidx];
 184         i_coord_offset   = DIM*inr;
 185
 186         /* Load i particle coords and add shift vector */
 187         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 188
 189         fix0             = _mm256_setzero_ps();
 190         fiy0             = _mm256_setzero_ps();
 191         fiz0             = _mm256_setzero_ps();
 192
 193         /* Load parameters for i particles */
 194         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 195         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 196
 197         /* Reset potential sums */
 198         velecsum         = _mm256_setzero_ps();
 199         vvdwsum          = _mm256_setzero_ps();
 200
 201         /* Start inner kernel loop */
 202         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 203         {
 204
 205             /* Get j neighbor index, and coordinate index */
 206             jnrA             = jjnr[jidx];
 207             jnrB             = jjnr[jidx+1];
 208             jnrC             = jjnr[jidx+2];
 209             jnrD             = jjnr[jidx+3];
 210             jnrE             = jjnr[jidx+4];
 211             jnrF             = jjnr[jidx+5];
 212             jnrG             = jjnr[jidx+6];
 213             jnrH             = jjnr[jidx+7];
 214             j_coord_offsetA  = DIM*jnrA;
 215             j_coord_offsetB  = DIM*jnrB;
 216             j_coord_offsetC  = DIM*jnrC;
 217             j_coord_offsetD  = DIM*jnrD;
 218             j_coord_offsetE  = DIM*jnrE;
 219             j_coord_offsetF  = DIM*jnrF;
 220             j_coord_offsetG  = DIM*jnrG;
 221             j_coord_offsetH  = DIM*jnrH;
 222
 223             /* load j atom coordinates */
 224             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 225                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 226                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 227                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 228                                                  &jx0,&jy0,&jz0);
 229
 230             /* Calculate displacement vector */
 231             dx00             = _mm256_sub_ps(ix0,jx0);
 232             dy00             = _mm256_sub_ps(iy0,jy0);
 233             dz00             = _mm256_sub_ps(iz0,jz0);
 234
 235             /* Calculate squared distance and things based on it */
 236             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 237
 238             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 239
 240             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 241
 242             /* Load parameters for j particles */
 243             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 244                                                                  charge+jnrC+0,charge+jnrD+0,
 245                                                                  charge+jnrE+0,charge+jnrF+0,
 246                                                                  charge+jnrG+0,charge+jnrH+0);
 247             vdwjidx0A        = 2*vdwtype[jnrA+0];
 248             vdwjidx0B        = 2*vdwtype[jnrB+0];
 249             vdwjidx0C        = 2*vdwtype[jnrC+0];
 250             vdwjidx0D        = 2*vdwtype[jnrD+0];
 251             vdwjidx0E        = 2*vdwtype[jnrE+0];
 252             vdwjidx0F        = 2*vdwtype[jnrF+0];
 253             vdwjidx0G        = 2*vdwtype[jnrG+0];
 254             vdwjidx0H        = 2*vdwtype[jnrH+0];
 255
 256             /**************************
 257              * CALCULATE INTERACTIONS *
 258              **************************/
 259
 260             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 261             {
 262
 263             r00              = _mm256_mul_ps(rsq00,rinv00);
 264
 265             /* Compute parameters for interactions between i and j atoms */
 266             qq00             = _mm256_mul_ps(iq0,jq0);
 267             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 268                                             vdwioffsetptr0+vdwjidx0B,
 269                                             vdwioffsetptr0+vdwjidx0C,
 270                                             vdwioffsetptr0+vdwjidx0D,
 271                                             vdwioffsetptr0+vdwjidx0E,
 272                                             vdwioffsetptr0+vdwjidx0F,
 273                                             vdwioffsetptr0+vdwjidx0G,
 274                                             vdwioffsetptr0+vdwjidx0H,
 275                                             &c6_00,&c12_00);
 276
 277             /* EWALD ELECTROSTATICS */
 278
 279             /* Analytical PME correction */
 280             zeta2            = _mm256_mul_ps(beta2,rsq00);
 281             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 282             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 283             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 284             felec            = _mm256_mul_ps(qq00,felec);
 285             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 286             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 287             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 288             velec            = _mm256_mul_ps(qq00,velec);
 289
 290             /* LENNARD-JONES DISPERSION/REPULSION */
 291
 292             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 293             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 294             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 295             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 296             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 297
 298             d                = _mm256_sub_ps(r00,rswitch);
 299             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 300             d2               = _mm256_mul_ps(d,d);
 301             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 302
 303             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 304
 305             /* Evaluate switch function */
 306             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 307             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 308             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 309             velec            = _mm256_mul_ps(velec,sw);
 310             vvdw             = _mm256_mul_ps(vvdw,sw);
 311             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 312
 313             /* Update potential sum for this i atom from the interaction with this j atom. */
 314             velec            = _mm256_and_ps(velec,cutoff_mask);
 315             velecsum         = _mm256_add_ps(velecsum,velec);
 316             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 317             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 318
 319             fscal            = _mm256_add_ps(felec,fvdw);
 320
 321             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 322
 323             /* Calculate temporary vectorial force */
 324             tx               = _mm256_mul_ps(fscal,dx00);
 325             ty               = _mm256_mul_ps(fscal,dy00);
 326             tz               = _mm256_mul_ps(fscal,dz00);
 327
 328             /* Update vectorial force */
 329             fix0             = _mm256_add_ps(fix0,tx);
 330             fiy0             = _mm256_add_ps(fiy0,ty);
 331             fiz0             = _mm256_add_ps(fiz0,tz);
 332
 333             fjptrA             = f+j_coord_offsetA;
 334             fjptrB             = f+j_coord_offsetB;
 335             fjptrC             = f+j_coord_offsetC;
 336             fjptrD             = f+j_coord_offsetD;
 337             fjptrE             = f+j_coord_offsetE;
 338             fjptrF             = f+j_coord_offsetF;
 339             fjptrG             = f+j_coord_offsetG;
 340             fjptrH             = f+j_coord_offsetH;
 341             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 342
 343             }
 344
 345             /* Inner loop uses 126 flops */
 346         }
 347
 348         if(jidx<j_index_end)
 349         {
 350
 351             /* Get j neighbor index, and coordinate index */
 352             jnrlistA         = jjnr[jidx];
 353             jnrlistB         = jjnr[jidx+1];
 354             jnrlistC         = jjnr[jidx+2];
 355             jnrlistD         = jjnr[jidx+3];
 356             jnrlistE         = jjnr[jidx+4];
 357             jnrlistF         = jjnr[jidx+5];
 358             jnrlistG         = jjnr[jidx+6];
 359             jnrlistH         = jjnr[jidx+7];
 360             /* Sign of each element will be negative for non-real atoms.
 361              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 362              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 363              */
 364             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 365                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 366
 367             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 368             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 369             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 370             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 371             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 372             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 373             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 374             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 375             j_coord_offsetA  = DIM*jnrA;
 376             j_coord_offsetB  = DIM*jnrB;
 377             j_coord_offsetC  = DIM*jnrC;
 378             j_coord_offsetD  = DIM*jnrD;
 379             j_coord_offsetE  = DIM*jnrE;
 380             j_coord_offsetF  = DIM*jnrF;
 381             j_coord_offsetG  = DIM*jnrG;
 382             j_coord_offsetH  = DIM*jnrH;
 383
 384             /* load j atom coordinates */
 385             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 386                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 387                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 388                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 389                                                  &jx0,&jy0,&jz0);
 390
 391             /* Calculate displacement vector */
 392             dx00             = _mm256_sub_ps(ix0,jx0);
 393             dy00             = _mm256_sub_ps(iy0,jy0);
 394             dz00             = _mm256_sub_ps(iz0,jz0);
 395
 396             /* Calculate squared distance and things based on it */
 397             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 398
 399             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 400
 401             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 402
 403             /* Load parameters for j particles */
 404             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 405                                                                  charge+jnrC+0,charge+jnrD+0,
 406                                                                  charge+jnrE+0,charge+jnrF+0,
 407                                                                  charge+jnrG+0,charge+jnrH+0);
 408             vdwjidx0A        = 2*vdwtype[jnrA+0];
 409             vdwjidx0B        = 2*vdwtype[jnrB+0];
 410             vdwjidx0C        = 2*vdwtype[jnrC+0];
 411             vdwjidx0D        = 2*vdwtype[jnrD+0];
 412             vdwjidx0E        = 2*vdwtype[jnrE+0];
 413             vdwjidx0F        = 2*vdwtype[jnrF+0];
 414             vdwjidx0G        = 2*vdwtype[jnrG+0];
 415             vdwjidx0H        = 2*vdwtype[jnrH+0];
 416
 417             /**************************
 418              * CALCULATE INTERACTIONS *
 419              **************************/
 420
 421             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 422             {
 423
 424             r00              = _mm256_mul_ps(rsq00,rinv00);
 425             r00              = _mm256_andnot_ps(dummy_mask,r00);
 426
 427             /* Compute parameters for interactions between i and j atoms */
 428             qq00             = _mm256_mul_ps(iq0,jq0);
 429             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 430                                             vdwioffsetptr0+vdwjidx0B,
 431                                             vdwioffsetptr0+vdwjidx0C,
 432                                             vdwioffsetptr0+vdwjidx0D,
 433                                             vdwioffsetptr0+vdwjidx0E,
 434                                             vdwioffsetptr0+vdwjidx0F,
 435                                             vdwioffsetptr0+vdwjidx0G,
 436                                             vdwioffsetptr0+vdwjidx0H,
 437                                             &c6_00,&c12_00);
 438
 439             /* EWALD ELECTROSTATICS */
 440
 441             /* Analytical PME correction */
 442             zeta2            = _mm256_mul_ps(beta2,rsq00);
 443             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 444             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 445             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 446             felec            = _mm256_mul_ps(qq00,felec);
 447             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 448             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 449             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 450             velec            = _mm256_mul_ps(qq00,velec);
 451
 452             /* LENNARD-JONES DISPERSION/REPULSION */
 453
 454             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 455             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 456             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 457             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 458             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 459
 460             d                = _mm256_sub_ps(r00,rswitch);
 461             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 462             d2               = _mm256_mul_ps(d,d);
 463             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 464
 465             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 466
 467             /* Evaluate switch function */
 468             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 469             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 470             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 471             velec            = _mm256_mul_ps(velec,sw);
 472             vvdw             = _mm256_mul_ps(vvdw,sw);
 473             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 474
 475             /* Update potential sum for this i atom from the interaction with this j atom. */
 476             velec            = _mm256_and_ps(velec,cutoff_mask);
 477             velec            = _mm256_andnot_ps(dummy_mask,velec);
 478             velecsum         = _mm256_add_ps(velecsum,velec);
 479             vvdw             = _mm256_and_ps(vvdw,cutoff_mask);
 480             vvdw             = _mm256_andnot_ps(dummy_mask,vvdw);
 481             vvdwsum          = _mm256_add_ps(vvdwsum,vvdw);
 482
 483             fscal            = _mm256_add_ps(felec,fvdw);
 484
 485             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 486
 487             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 488
 489             /* Calculate temporary vectorial force */
 490             tx               = _mm256_mul_ps(fscal,dx00);
 491             ty               = _mm256_mul_ps(fscal,dy00);
 492             tz               = _mm256_mul_ps(fscal,dz00);
 493
 494             /* Update vectorial force */
 495             fix0             = _mm256_add_ps(fix0,tx);
 496             fiy0             = _mm256_add_ps(fiy0,ty);
 497             fiz0             = _mm256_add_ps(fiz0,tz);
 498
 499             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 500             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 501             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 502             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 503             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 504             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 505             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 506             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 507             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 508
 509             }
 510
 511             /* Inner loop uses 127 flops */
 512         }
 513
 514         /* End of innermost loop */
 515
 516         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 517                                                  f+i_coord_offset,fshift+i_shift_offset);
 518
 519         ggid                        = gid[iidx];
 520         /* Update potential energies */
 521         gmx_mm256_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 522         gmx_mm256_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 523
 524         /* Increment number of inner iterations */
 525         inneriter                  += j_index_end - j_index_start;
 526
 527         /* Outer loop uses 9 flops */
 528     }
 529
 530     /* Increment number of outer iterations */
 531     outeriter        += nri;
 532
 533     /* Update outer/inner flops */
 534
 535     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_VF,outeriter*9 + inneriter*127);
 536 }
 537 /*
 538  * Gromacs nonbonded kernel:   nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_avx_256_single
 539  * Electrostatics interaction: Ewald
 540  * VdW interaction:            LennardJones
 541  * Geometry:                   Particle-Particle
 542  * Calculate force/pot:        Force
 543  */
 544 void
 545 nb_kernel_ElecEwSw_VdwLJSw_GeomP1P1_F_avx_256_single
 546                     (t_nblist                    * gmx_restrict       nlist,
 547                      rvec                        * gmx_restrict          xx,
 548                      rvec                        * gmx_restrict          ff,
 549                      t_forcerec                  * gmx_restrict          fr,
 550                      t_mdatoms                   * gmx_restrict     mdatoms,
 551                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 552                      t_nrnb                      * gmx_restrict        nrnb)
 553 {
 554     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 555      * just 0 for non-waters.
 556      * Suffixes A,B,C,D,E,F,G,H refer to j loop unrolling done with AVX, e.g. for the eight different
 557      * jnr indices corresponding to data put in the four positions in the SIMD register.
 558      */
 559     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 560     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 561     int              jnrA,jnrB,jnrC,jnrD;
 562     int              jnrE,jnrF,jnrG,jnrH;
 563     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 564     int              jnrlistE,jnrlistF,jnrlistG,jnrlistH;
 565     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 566     int              j_coord_offsetE,j_coord_offsetF,j_coord_offsetG,j_coord_offsetH;
 567     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 568     real             rcutoff_scalar;
 569     real             *shiftvec,*fshift,*x,*f;
 570     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD,*fjptrE,*fjptrF,*fjptrG,*fjptrH;
 571     real             scratch[4*DIM];
 572     __m256           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 573     real *           vdwioffsetptr0;
 574     __m256           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 575     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D,vdwjidx0E,vdwjidx0F,vdwjidx0G,vdwjidx0H;
 576     __m256           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 577     __m256           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 578     __m256           velec,felec,velecsum,facel,crf,krf,krf2;
 579     real             *charge;
 580     int              nvdwtype;
 581     __m256           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 582     int              *vdwtype;
 583     real             *vdwparam;
 584     __m256           one_sixth   = _mm256_set1_ps(1.0/6.0);
 585     __m256           one_twelfth = _mm256_set1_ps(1.0/12.0);
 586     __m256i          ewitab;
 587     __m128i          ewitab_lo,ewitab_hi;
 588     __m256           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 589     __m256           beta,beta2,beta3,zeta2,pmecorrF,pmecorrV,rinv3;
 590     real             *ewtab;
 591     __m256           rswitch,swV3,swV4,swV5,swF2,swF3,swF4,d,d2,sw,dsw;
 592     real             rswitch_scalar,d_scalar;
 593     __m256           dummy_mask,cutoff_mask;
 594     __m256           signbit = _mm256_castsi256_ps( _mm256_set1_epi32(0x80000000) );
 595     __m256           one     = _mm256_set1_ps(1.0);
 596     __m256           two     = _mm256_set1_ps(2.0);
 597     x                = xx[0];
 598     f                = ff[0];
 599
 600     nri              = nlist->nri;
 601     iinr             = nlist->iinr;
 602     jindex           = nlist->jindex;
 603     jjnr             = nlist->jjnr;
 604     shiftidx         = nlist->shift;
 605     gid              = nlist->gid;
 606     shiftvec         = fr->shift_vec[0];
 607     fshift           = fr->fshift[0];
 608     facel            = _mm256_set1_ps(fr->epsfac);
 609     charge           = mdatoms->chargeA;
 610     nvdwtype         = fr->ntype;
 611     vdwparam         = fr->nbfp;
 612     vdwtype          = mdatoms->typeA;
 613
 614     sh_ewald         = _mm256_set1_ps(fr->ic->sh_ewald);
 615     beta             = _mm256_set1_ps(fr->ic->ewaldcoeff_q);
 616     beta2            = _mm256_mul_ps(beta,beta);
 617     beta3            = _mm256_mul_ps(beta,beta2);
 618
 619     ewtab            = fr->ic->tabq_coul_FDV0;
 620     ewtabscale       = _mm256_set1_ps(fr->ic->tabq_scale);
 621     ewtabhalfspace   = _mm256_set1_ps(0.5/fr->ic->tabq_scale);
 622
 623     /* When we use explicit cutoffs the value must be identical for elec and VdW, so use elec as an arbitrary choice */
 624     rcutoff_scalar   = fr->rcoulomb;
 625     rcutoff          = _mm256_set1_ps(rcutoff_scalar);
 626     rcutoff2         = _mm256_mul_ps(rcutoff,rcutoff);
 627
 628     rswitch_scalar   = fr->rcoulomb_switch;
 629     rswitch          = _mm256_set1_ps(rswitch_scalar);
 630     /* Setup switch parameters */
 631     d_scalar         = rcutoff_scalar-rswitch_scalar;
 632     d                = _mm256_set1_ps(d_scalar);
 633     swV3             = _mm256_set1_ps(-10.0/(d_scalar*d_scalar*d_scalar));
 634     swV4             = _mm256_set1_ps( 15.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 635     swV5             = _mm256_set1_ps( -6.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 636     swF2             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar));
 637     swF3             = _mm256_set1_ps( 60.0/(d_scalar*d_scalar*d_scalar*d_scalar));
 638     swF4             = _mm256_set1_ps(-30.0/(d_scalar*d_scalar*d_scalar*d_scalar*d_scalar));
 639
 640     /* Avoid stupid compiler warnings */
 641     jnrA = jnrB = jnrC = jnrD = jnrE = jnrF = jnrG = jnrH = 0;
 642     j_coord_offsetA = 0;
 643     j_coord_offsetB = 0;
 644     j_coord_offsetC = 0;
 645     j_coord_offsetD = 0;
 646     j_coord_offsetE = 0;
 647     j_coord_offsetF = 0;
 648     j_coord_offsetG = 0;
 649     j_coord_offsetH = 0;
 650
 651     outeriter        = 0;
 652     inneriter        = 0;
 653
 654     for(iidx=0;iidx<4*DIM;iidx++)
 655     {
 656         scratch[iidx] = 0.0;
 657     }
 658
 659     /* Start outer loop over neighborlists */
 660     for(iidx=0; iidx<nri; iidx++)
 661     {
 662         /* Load shift vector for this list */
 663         i_shift_offset   = DIM*shiftidx[iidx];
 664
 665         /* Load limits for loop over neighbors */
 666         j_index_start    = jindex[iidx];
 667         j_index_end      = jindex[iidx+1];
 668
 669         /* Get outer coordinate index */
 670         inr              = iinr[iidx];
 671         i_coord_offset   = DIM*inr;
 672
 673         /* Load i particle coords and add shift vector */
 674         gmx_mm256_load_shift_and_1rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,&ix0,&iy0,&iz0);
 675
 676         fix0             = _mm256_setzero_ps();
 677         fiy0             = _mm256_setzero_ps();
 678         fiz0             = _mm256_setzero_ps();
 679
 680         /* Load parameters for i particles */
 681         iq0              = _mm256_mul_ps(facel,_mm256_set1_ps(charge[inr+0]));
 682         vdwioffsetptr0   = vdwparam+2*nvdwtype*vdwtype[inr+0];
 683
 684         /* Start inner kernel loop */
 685         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+7]>=0; jidx+=8)
 686         {
 687
 688             /* Get j neighbor index, and coordinate index */
 689             jnrA             = jjnr[jidx];
 690             jnrB             = jjnr[jidx+1];
 691             jnrC             = jjnr[jidx+2];
 692             jnrD             = jjnr[jidx+3];
 693             jnrE             = jjnr[jidx+4];
 694             jnrF             = jjnr[jidx+5];
 695             jnrG             = jjnr[jidx+6];
 696             jnrH             = jjnr[jidx+7];
 697             j_coord_offsetA  = DIM*jnrA;
 698             j_coord_offsetB  = DIM*jnrB;
 699             j_coord_offsetC  = DIM*jnrC;
 700             j_coord_offsetD  = DIM*jnrD;
 701             j_coord_offsetE  = DIM*jnrE;
 702             j_coord_offsetF  = DIM*jnrF;
 703             j_coord_offsetG  = DIM*jnrG;
 704             j_coord_offsetH  = DIM*jnrH;
 705
 706             /* load j atom coordinates */
 707             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 708                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 709                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 710                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 711                                                  &jx0,&jy0,&jz0);
 712
 713             /* Calculate displacement vector */
 714             dx00             = _mm256_sub_ps(ix0,jx0);
 715             dy00             = _mm256_sub_ps(iy0,jy0);
 716             dz00             = _mm256_sub_ps(iz0,jz0);
 717
 718             /* Calculate squared distance and things based on it */
 719             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 720
 721             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 722
 723             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 724
 725             /* Load parameters for j particles */
 726             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 727                                                                  charge+jnrC+0,charge+jnrD+0,
 728                                                                  charge+jnrE+0,charge+jnrF+0,
 729                                                                  charge+jnrG+0,charge+jnrH+0);
 730             vdwjidx0A        = 2*vdwtype[jnrA+0];
 731             vdwjidx0B        = 2*vdwtype[jnrB+0];
 732             vdwjidx0C        = 2*vdwtype[jnrC+0];
 733             vdwjidx0D        = 2*vdwtype[jnrD+0];
 734             vdwjidx0E        = 2*vdwtype[jnrE+0];
 735             vdwjidx0F        = 2*vdwtype[jnrF+0];
 736             vdwjidx0G        = 2*vdwtype[jnrG+0];
 737             vdwjidx0H        = 2*vdwtype[jnrH+0];
 738
 739             /**************************
 740              * CALCULATE INTERACTIONS *
 741              **************************/
 742
 743             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 744             {
 745
 746             r00              = _mm256_mul_ps(rsq00,rinv00);
 747
 748             /* Compute parameters for interactions between i and j atoms */
 749             qq00             = _mm256_mul_ps(iq0,jq0);
 750             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 751                                             vdwioffsetptr0+vdwjidx0B,
 752                                             vdwioffsetptr0+vdwjidx0C,
 753                                             vdwioffsetptr0+vdwjidx0D,
 754                                             vdwioffsetptr0+vdwjidx0E,
 755                                             vdwioffsetptr0+vdwjidx0F,
 756                                             vdwioffsetptr0+vdwjidx0G,
 757                                             vdwioffsetptr0+vdwjidx0H,
 758                                             &c6_00,&c12_00);
 759
 760             /* EWALD ELECTROSTATICS */
 761
 762             /* Analytical PME correction */
 763             zeta2            = _mm256_mul_ps(beta2,rsq00);
 764             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 765             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 766             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 767             felec            = _mm256_mul_ps(qq00,felec);
 768             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 769             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 770             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 771             velec            = _mm256_mul_ps(qq00,velec);
 772
 773             /* LENNARD-JONES DISPERSION/REPULSION */
 774
 775             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 776             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 777             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 778             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 779             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 780
 781             d                = _mm256_sub_ps(r00,rswitch);
 782             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 783             d2               = _mm256_mul_ps(d,d);
 784             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 785
 786             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 787
 788             /* Evaluate switch function */
 789             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 790             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 791             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 792             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 793
 794             fscal            = _mm256_add_ps(felec,fvdw);
 795
 796             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 797
 798             /* Calculate temporary vectorial force */
 799             tx               = _mm256_mul_ps(fscal,dx00);
 800             ty               = _mm256_mul_ps(fscal,dy00);
 801             tz               = _mm256_mul_ps(fscal,dz00);
 802
 803             /* Update vectorial force */
 804             fix0             = _mm256_add_ps(fix0,tx);
 805             fiy0             = _mm256_add_ps(fiy0,ty);
 806             fiz0             = _mm256_add_ps(fiz0,tz);
 807
 808             fjptrA             = f+j_coord_offsetA;
 809             fjptrB             = f+j_coord_offsetB;
 810             fjptrC             = f+j_coord_offsetC;
 811             fjptrD             = f+j_coord_offsetD;
 812             fjptrE             = f+j_coord_offsetE;
 813             fjptrF             = f+j_coord_offsetF;
 814             fjptrG             = f+j_coord_offsetG;
 815             fjptrH             = f+j_coord_offsetH;
 816             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 817
 818             }
 819
 820             /* Inner loop uses 120 flops */
 821         }
 822
 823         if(jidx<j_index_end)
 824         {
 825
 826             /* Get j neighbor index, and coordinate index */
 827             jnrlistA         = jjnr[jidx];
 828             jnrlistB         = jjnr[jidx+1];
 829             jnrlistC         = jjnr[jidx+2];
 830             jnrlistD         = jjnr[jidx+3];
 831             jnrlistE         = jjnr[jidx+4];
 832             jnrlistF         = jjnr[jidx+5];
 833             jnrlistG         = jjnr[jidx+6];
 834             jnrlistH         = jjnr[jidx+7];
 835             /* Sign of each element will be negative for non-real atoms.
 836              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 837              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 838              */
 839             dummy_mask = gmx_mm256_set_m128(gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx+4)),_mm_setzero_si128())),
 840                                             gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128())));
 841
 842             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 843             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 844             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 845             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 846             jnrE       = (jnrlistE>=0) ? jnrlistE : 0;
 847             jnrF       = (jnrlistF>=0) ? jnrlistF : 0;
 848             jnrG       = (jnrlistG>=0) ? jnrlistG : 0;
 849             jnrH       = (jnrlistH>=0) ? jnrlistH : 0;
 850             j_coord_offsetA  = DIM*jnrA;
 851             j_coord_offsetB  = DIM*jnrB;
 852             j_coord_offsetC  = DIM*jnrC;
 853             j_coord_offsetD  = DIM*jnrD;
 854             j_coord_offsetE  = DIM*jnrE;
 855             j_coord_offsetF  = DIM*jnrF;
 856             j_coord_offsetG  = DIM*jnrG;
 857             j_coord_offsetH  = DIM*jnrH;
 858
 859             /* load j atom coordinates */
 860             gmx_mm256_load_1rvec_8ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 861                                                  x+j_coord_offsetC,x+j_coord_offsetD,
 862                                                  x+j_coord_offsetE,x+j_coord_offsetF,
 863                                                  x+j_coord_offsetG,x+j_coord_offsetH,
 864                                                  &jx0,&jy0,&jz0);
 865
 866             /* Calculate displacement vector */
 867             dx00             = _mm256_sub_ps(ix0,jx0);
 868             dy00             = _mm256_sub_ps(iy0,jy0);
 869             dz00             = _mm256_sub_ps(iz0,jz0);
 870
 871             /* Calculate squared distance and things based on it */
 872             rsq00            = gmx_mm256_calc_rsq_ps(dx00,dy00,dz00);
 873
 874             rinv00           = gmx_mm256_invsqrt_ps(rsq00);
 875
 876             rinvsq00         = _mm256_mul_ps(rinv00,rinv00);
 877
 878             /* Load parameters for j particles */
 879             jq0              = gmx_mm256_load_8real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 880                                                                  charge+jnrC+0,charge+jnrD+0,
 881                                                                  charge+jnrE+0,charge+jnrF+0,
 882                                                                  charge+jnrG+0,charge+jnrH+0);
 883             vdwjidx0A        = 2*vdwtype[jnrA+0];
 884             vdwjidx0B        = 2*vdwtype[jnrB+0];
 885             vdwjidx0C        = 2*vdwtype[jnrC+0];
 886             vdwjidx0D        = 2*vdwtype[jnrD+0];
 887             vdwjidx0E        = 2*vdwtype[jnrE+0];
 888             vdwjidx0F        = 2*vdwtype[jnrF+0];
 889             vdwjidx0G        = 2*vdwtype[jnrG+0];
 890             vdwjidx0H        = 2*vdwtype[jnrH+0];
 891
 892             /**************************
 893              * CALCULATE INTERACTIONS *
 894              **************************/
 895
 896             if (gmx_mm256_any_lt(rsq00,rcutoff2))
 897             {
 898
 899             r00              = _mm256_mul_ps(rsq00,rinv00);
 900             r00              = _mm256_andnot_ps(dummy_mask,r00);
 901
 902             /* Compute parameters for interactions between i and j atoms */
 903             qq00             = _mm256_mul_ps(iq0,jq0);
 904             gmx_mm256_load_8pair_swizzle_ps(vdwioffsetptr0+vdwjidx0A,
 905                                             vdwioffsetptr0+vdwjidx0B,
 906                                             vdwioffsetptr0+vdwjidx0C,
 907                                             vdwioffsetptr0+vdwjidx0D,
 908                                             vdwioffsetptr0+vdwjidx0E,
 909                                             vdwioffsetptr0+vdwjidx0F,
 910                                             vdwioffsetptr0+vdwjidx0G,
 911                                             vdwioffsetptr0+vdwjidx0H,
 912                                             &c6_00,&c12_00);
 913
 914             /* EWALD ELECTROSTATICS */
 915
 916             /* Analytical PME correction */
 917             zeta2            = _mm256_mul_ps(beta2,rsq00);
 918             rinv3            = _mm256_mul_ps(rinvsq00,rinv00);
 919             pmecorrF         = gmx_mm256_pmecorrF_ps(zeta2);
 920             felec            = _mm256_add_ps( _mm256_mul_ps(pmecorrF,beta3), rinv3);
 921             felec            = _mm256_mul_ps(qq00,felec);
 922             pmecorrV         = gmx_mm256_pmecorrV_ps(zeta2);
 923             pmecorrV         = _mm256_mul_ps(pmecorrV,beta);
 924             velec            = _mm256_sub_ps(rinv00,pmecorrV);
 925             velec            = _mm256_mul_ps(qq00,velec);
 926
 927             /* LENNARD-JONES DISPERSION/REPULSION */
 928
 929             rinvsix          = _mm256_mul_ps(_mm256_mul_ps(rinvsq00,rinvsq00),rinvsq00);
 930             vvdw6            = _mm256_mul_ps(c6_00,rinvsix);
 931             vvdw12           = _mm256_mul_ps(c12_00,_mm256_mul_ps(rinvsix,rinvsix));
 932             vvdw             = _mm256_sub_ps( _mm256_mul_ps(vvdw12,one_twelfth) , _mm256_mul_ps(vvdw6,one_sixth) );
 933             fvdw             = _mm256_mul_ps(_mm256_sub_ps(vvdw12,vvdw6),rinvsq00);
 934
 935             d                = _mm256_sub_ps(r00,rswitch);
 936             d                = _mm256_max_ps(d,_mm256_setzero_ps());
 937             d2               = _mm256_mul_ps(d,d);
 938             sw               = _mm256_add_ps(one,_mm256_mul_ps(d2,_mm256_mul_ps(d,_mm256_add_ps(swV3,_mm256_mul_ps(d,_mm256_add_ps(swV4,_mm256_mul_ps(d,swV5)))))));
 939
 940             dsw              = _mm256_mul_ps(d2,_mm256_add_ps(swF2,_mm256_mul_ps(d,_mm256_add_ps(swF3,_mm256_mul_ps(d,swF4)))));
 941
 942             /* Evaluate switch function */
 943             /* fscal'=f'/r=-(v*sw)'/r=-(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=fscal*sw-v*dsw/r */
 944             felec            = _mm256_sub_ps( _mm256_mul_ps(felec,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(velec,dsw)) );
 945             fvdw             = _mm256_sub_ps( _mm256_mul_ps(fvdw,sw) , _mm256_mul_ps(rinv00,_mm256_mul_ps(vvdw,dsw)) );
 946             cutoff_mask      = _mm256_cmp_ps(rsq00,rcutoff2,_CMP_LT_OQ);
 947
 948             fscal            = _mm256_add_ps(felec,fvdw);
 949
 950             fscal            = _mm256_and_ps(fscal,cutoff_mask);
 951
 952             fscal            = _mm256_andnot_ps(dummy_mask,fscal);
 953
 954             /* Calculate temporary vectorial force */
 955             tx               = _mm256_mul_ps(fscal,dx00);
 956             ty               = _mm256_mul_ps(fscal,dy00);
 957             tz               = _mm256_mul_ps(fscal,dz00);
 958
 959             /* Update vectorial force */
 960             fix0             = _mm256_add_ps(fix0,tx);
 961             fiy0             = _mm256_add_ps(fiy0,ty);
 962             fiz0             = _mm256_add_ps(fiz0,tz);
 963
 964             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 965             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 966             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 967             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 968             fjptrE             = (jnrlistE>=0) ? f+j_coord_offsetE : scratch;
 969             fjptrF             = (jnrlistF>=0) ? f+j_coord_offsetF : scratch;
 970             fjptrG             = (jnrlistG>=0) ? f+j_coord_offsetG : scratch;
 971             fjptrH             = (jnrlistH>=0) ? f+j_coord_offsetH : scratch;
 972             gmx_mm256_decrement_1rvec_8ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjptrE,fjptrF,fjptrG,fjptrH,tx,ty,tz);
 973
 974             }
 975
 976             /* Inner loop uses 121 flops */
 977         }
 978
 979         /* End of innermost loop */
 980
 981         gmx_mm256_update_iforce_1atom_swizzle_ps(fix0,fiy0,fiz0,
 982                                                  f+i_coord_offset,fshift+i_shift_offset);
 983
 984         /* Increment number of inner iterations */
 985         inneriter                  += j_index_end - j_index_start;
 986
 987         /* Outer loop uses 7 flops */
 988     }
 989
 990     /* Increment number of outer iterations */
 991     outeriter        += nri;
 992
 993     /* Update outer/inner flops */
 994
 995     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_F,outeriter*7 + inneriter*121);
 996 }